An Overview of Myrinet By: Ralph Zajac What is Myrinet? LAN designed for clusters Based on USCD’s ATOMIC LAN Has many characteristics of MPP message-passing networks MPP Network Characteristics High data rate Regular topology and scalability Very low error rate Cut-through routing Flow control ATOMIC LAN: Overview Based on technology from Caltech’s Mosaic project Used TCP/IP at burst rates of 400 Mb/s Error rate less than 10-15 ATOMIC LAN: Limitations Asynchronous communication could lead to deadlock Network topology – Complex – Hidden computing power Performance limited by lack of DMA in host interface End-to-end data rates limited by TCP/IP stack of OS Myrinet Designed to address the limitations of ATOMIC LAN Not limited to an existing MPP network Specifications published in ANSI/VITA 26-1998 Myrinet: Links Can be copper (up to 25m) or opticalfiber Full duplex Uses non-return-tozero encoding Flow control run by a slack buffer STOP GO Slack Buffer Myrinet: Packets and Routing Leading byte determines the outgoing port MSB of each header byte distinguishes between “host” and “switch” packets Variable payload length 8-bit CRC Myrinet: Switches Use blocking-cut-through routing Each contains two custom VLSI chips – crossbar-switch – dual interface Topology is arbitrary Myrinet: Host Interface Available for PCI and Sbus 32-bit SRAM for MCP Can act as a slave or a bus master Myrinet: Software MCP – Handles all interfacing with the network – Continuously mapping and monitoring the network Host Software – TCP/IP and UDP/IP – Myrinet API High Performance Virtual Machine Windows NT 4.0 cluster built at UCSD and UIUC for the NCSA 96 dual processor nodes Used a fat-tree network with 8-port Myrinet switches Was a 56 GFlops system with: – 45 GB of DRAM – 1.6 MB/s bisection bandwidth – 400 GB of disk storage HPVM: Testing Tested on four programs – ZeusMP: astrophysical phenomena in 3D – Cactus: modular 3D numerical relativity – AS-PCG kernel: linear systems solver – QMC kernel: Quantum Monte Carlo simulator Compared with a Cray T3E and an Origin 2000 HPVM: Results Performed from 2 to 4 times slower – Slower floating point calculations – Flow control limitations of Myrinet Much better price/performance ratio Conclusions Myrinet is: – Flexible – Fast Affordable systems with acceptable performance